美国政府首次关停AI模型，去中心化AI获验证

On June 12th, the US government did something it had never done before. It reached into a live deployed AI model and switched it off. At precisely 5:21 p.m. Eastern on a Friday afternoon, the Commerce Department's Bureau of Industry and Security sent Anthropic CEO Dario Amodei a formal export control directive signed by Commerce Secretary Howard Lutnick with explicit warnings of civil and criminal penalties for non-compliance. [music] Just 90 minutes later, two of Anthropic's frontier models, Fable 5 and Mythos 5, went dark worldwide.

And as I record this, they're still [music] offline. It didn't take a hack, a crash, or a courtroom order. A simple letter brought the models down. So, today I'm going to lay out exactly what happened on that Friday, show you why this is the real argument for decentralized AI that nobody framed properly until now, and then steelman the skeptics who think the whole decentralized AI narrative is just a pump wearing a nicer suit.

My name is DC and you're watching The Coin Bureau. Now, before we get into crypto at all, you need to understand the climate this landed in because timing here is everything. Bitcoin is sitting at around $63,700 as I record this video. The broader market is jittery, and AI has been the one narrative hoovering up capital where everything else bleeds. Into that environment, the US government just proved that the most valuable AI lab on the planet, a company with a $47 billion run rate that filed for an APO 11 days before this happened, can have its flagship product turned off by a single agency.

And that is the backdrop that sent money racing into a corner of crypto most people had written off entirely. So, let's start with what actually happened because the mechanics matter. Fable 5 and Mythos 5 launched on June 9th. They were live for exactly 72 hours before Washington pulled the plug. The legal authority cited was an export administration regulations and specifically something called the deemed export rule under the Export Controls Reform Act of 2018.

Now, for those unfamiliar, let me put that simply. A deemed export means that sharing control technology with any foreign national, even one standing inside the United States, legally counts as exporting it. This is the same framework used to control advanced semiconductors, encryption, and nuclear materials. And on June 12th, for the very first time, the government applied it to a commercial chatbot. The stated reason was a vulnerability, a so-called narrow non-universal jailbreak, that could let the models help identify software vulnerabilities, which Washington feared could accelerate offensive cyber operations by adversaries like China.

And here's the detail that's almost too perfect. The jailbreak was reportedly discovered by Amazon researchers, and Amazon CEO Andy Jassy is the one who flagged it to the administration. The catch? Amazon's own enterprise service, Bedrock, runs Claude integrations. And when the models went dark, Bedrock customers got disrupted, too. So, in other words, the man who pulled the fire alarm set off the sprinklers in his own building.

Okay, so why did this become a global blackout instead of a targeted block? Because the directive demanded Anthropic cut off all foreign nationals, including its own foreign employees. And Anthropic argued it's simply not technically possible to verify a user's citizenship in real time across its entire API at scale. So, faced with partial compliance, which was legally radioactive, or a full shutdown, the company killed both models worldwide within 90 minutes.

Which meant American citizens got cut off, too. A foreign access order became a total product recall. Now, you might assume this is just a one-off regulatory thing that gets resolved in a week. Well, not really. As of mid-June, negotiations with the White House had failed to produce any resolution, or even an exemption for close allies like the UK. And that brings us to the deeper layer, the part that should make every person in crypto sit up.

June 12th exposed a permanent structural vulnerability in how these models operate. For years, centralized AI providers have built kill switches into their products by design. OpenAI, Anthropic, and Google already block API access by country to comply with sanctions on places like Iran, North Korea, and Russia. Italy's regulator temporarily banned ChatGPT back in 2023. China built an entirely separate walled-off AI stack because it couldn't trust access to American models.

So, the architecture for switching AI off has always existed, but every previous case was a geographic filter, a block by location. June 12th was something new entirely. It was a citizenship-based extraterritorial order, a government demanding that a company cut off anyone who isn't American anywhere on Earth. Analysts at Infotech Research called this sovereignty climbing the stack. So, to put it simply, governments have escalated beyond fighting over the data and chips to asserting direct control over the intelligence layer itself.

And the people in charge have not been subtle about it. Former UK Security Minister Tom Tugendhat said disabling these models wasn't a mistake, but the inevitable result of technology shaping warfare so that sovereignty is more about code than cannons. Canadian Prime Minister Mark Carney compared the moment to the 2008 financial crisis, warning that over-reliance on a handful of centralized AI providers creates systemic risk for entire national economies.

Even FIRE, the free speech group, called it an arbitrary abuse of power, comparing it to the kill switch tactics of regressive regimes. So, make no mistake, this is the censorship-resistant case for decentralized AI, and it just got a live proof point. Because here's the structural difference. A network like Bittensor has no Dario Amodei to send a letter to. It's a decentralized blockchain running an open marketplace for machine intelligence across 128 plus globally distributed subnets, up significantly from where it stood a year ago.

Participants contribute compute and models and earn Tao tokens in return. Shutting it down requires a simultaneous attack on thousands of node operators scattered across every jurisdiction on the planet. There is no point of failure for a government to press, and a project like Venice with VVV takes the same logic to private uncensored inference, positioning itself as AI that literally cannot be switched off by order.

Which brings us directly to where the money went. In the week after June 12th, AI crypto assets saw significant inflows as investors rotated into decentralized infrastructure plays with Tao leading the charge. Tao jumped somewhere between 28 and 30% within 12 hours, briefly touching a 3-week high near $283. Venice's VVV also reportedly gained in the same period, though independent price data for VVV is difficult to verify at this time.

Grayscale's head of research, Zach Pandl, came out and explicitly framed the regulatory squeeze as validating the need for censorship-resistant decentralized infrastructure. Now, I have to be straight with you here. This is exactly the kind of reflexive fear-driven rotation we've seen a dozen of times in our industry. The market priced in decentralized AI's potential value before that value has actually been proven. So, I'm not going to celebrate a green candle because a single 30% candle proves absolutely nothing.

What I will defend is the fundamentals you can actually point to. Bittensor has reportedly generated real on-chain revenue across its subnets with figures circulating around $43 million for Q1 2026. Though, that's a number worth verifying directly against on-chain data. These are real fees paid for actual AI computation delivered across its subnets, completely separate from token inflation or white paper promises. Now, let me give you the honest math on that.

If that figure holds, $43 million a quarter annualizes to roughly $172 million against a $2.55 bill

原文超出正文长度上限，此处截断——上游还有内容，完整版见上方「原文 ↗」。